Publications
Publication Information
| Title | Performance Advances in Computing Disconnected Diagrams with Deflation |
| Authors | Eloy Romero Alcalde, Andreas Stathopoulos, Konstantinos Orginos |
| JLAB number | JLAB-CST-21-3501 |
| LANL number | (None) |
| Other number | DOE/OR/23177-5319 |
| Document Type(s) | (Meeting) |
| Associated with EIC: | No |
| Supported by Jefferson Lab LDRD Funding: | No |
| Funding Source: | Nuclear Physics (NP) |
| Other Funding: | Exascale Computing Project 17-SC-20-SC |
|
Meeting Poster compiled for ECPAM 2021 | |
| Publication Abstract: | In this work, we propose two approaches to improve the performance scalability of deflation for variance reduction in stochastic trace estimation of large matrix inverses. These inverses stem from computing the disconnected contributions of correlation functions in lattice QCD. We reduce the cost of computing and applying deflation by using a multigrid prolongator as the deflation basis. We also improve the scalability of the multigrid solver by taking advantage of the simultaneous resolution of several linear systems. |
| Experiment Numbers: | other |
| Group: | Scientific Computing |
| Document: | |
| DOI: | |
| Accepted Manuscript: | |
| Supporting Documents: | |
| Supporting Datasets: | |